Pivotal MRMC Study of CAD Software With 14,400 Reviews With and Without CAD

Challenge

A pivotal multi-reader multi-case study was required to assess the clinical performance of computer-aided detection software, with more than 14,400 reviews to be performed without CAD and then with CAD. The sponsor's claim depended on demonstrating that readers detected more true findings with CAD without an unacceptable increase in false positives, which requires precise control over how and when readers see CAD marks.

With a review volume of this size, the study also needed medical guidance on case selection and reference-standard definition, a reader panel large enough to complete the work within the sponsor's timeline, and quality control that could detect protocol deviations such as incomplete unaided reads before they affected the analysis.

Intrinsic Imaging’s Solution

Intrinsic Imaging provided medical guidance on the case set, defining inclusion criteria, case-mix targets and the reference-standard methodology, and assembled the cases against a truth established by an independent panel with documented adjudication.

All case reviewers were provided by Intrinsic Imaging and trained on the reading protocol, which specified unaided interpretation followed by CAD-assisted interpretation under the sequential design appropriate to the device's intended use, with randomized case order and a validated reading platform recording each mark, confidence score and the exact sequence of unaided and aided decisions.

Quality control tracked completion of both reading phases per case and reader, flagged deviations for correction before the next session, and the final dataset was locked and delivered for the sponsor's MRMC statistical analysis within the quality framework of the company's five ISO certifications.

Intrinsic Imaging’s Results

• More than 14,400 reviews completed without and with CAD by a 24-reader panel within the sponsor's timeline
• Sequential unaided-then-aided reading enforced and recorded per case, with 100% of aided reads preceded by a completed unaided read
• Reference standard established by an independent truth panel with documented adjudication for every discordant case
• Dataset delivered in MRMC analysis format with full audit trail, supporting the sponsor's FDA submission